Dataset · adyen
Multi-step reasoning benchmark for data analysis AI agents
adyen/DABstep
Large text benchmark DABstep by Adyen for evaluating AI agents on multi-step reasoning tasks.
What's inside
The dataset contains from 1 to 10 million text entries and is divided into three splits: tasks, submissions, and task_scores. It was created by Adyen to support the benchmark and leaderboard of the same name, Data Agent Benchmark for Multi-step Reasoning. The tasks split is used to run and verify models, while the other two serve the leaderboard table.
What you can do with it
- Testing and comparing AI agents for data analysis on multi-step tasks
- Validating reasoning pipelines before submitting solutions to the DABstep leaderboard
- Size
- 1M–10M rows
- Downloads this month
- 157.5K
- Likes
- 61
How to load it
from datasets import load_dataset
ds = load_dataset("adyen/DABstep")